The ADJUST=DUNNETT DIFF=CONTROL(...) pairwise comparisons may not converge when used in a k-sample tests scenario on the STRATA statement (when the GROUP= option is NOT specified). A WARNING message is written to the log and the adjusted pvalues are reported as missing values. This problem affects all k-sample tests such as the logrank, Wilcoxon, Tarone, Peto, ModPeto, Fleming and LR tests.

Currently, a WARNING message is written to the log:

WARNING: The ADJUST=DUNNETT p-values and confidence limits 
              cannot be computed because the Dunnett-Hsu 
              approximation did not converge.  Try 
              ADJUST=SIMULATE.

The Dunnett-Hsu approximation depends on an iterative algorithm to approximate the covariance matrix. In the k-sample tests scenario, the covariance matrix is singular, which causes problems with the iterative algorithm.

To circumvent the problem, use ADJUST=SIMULATE.

Operating System and Release Information